Automatic Severity Classification of Diabetic Retinopathy Based on DenseNet and Convolutional Block Attention Module (original) (raw)
Related papers
An automated severity classification model for diabetic retinopathy
IRJET, 2023
Diabetic Retinopathy (DR)-a complication developed due to heightened blood glucose levels-is deemed one of the most sight-threatening diseases. Unfortunately, an ophthalmologist, a process that can be considered incorrect and time consuming, manually acquires the DR screening. In view of the huge increase in diabetic patients over recent years, automated diagnostic tests for diabetes have also become an increasingly important research topic. Additionally, Convolutional Neural Networks (CNN) have proven themselves to be state-of-the-art for DR stage diagnosis in recent times. This study offers a fresh, automatically learning-based method for determining severity from a single Colour Fundus Photograph (CFP). The suggested method builds a visual embedding using DenseNet169's encoder.. Convolutional Block Attention Module (CBAM) is also added on top of the encoder to boost its ability to discriminate. The model is then trained using the Kaggle Asia Pacific Tele-Ophthalmology Society (APTOS) dataset using cross-entropy loss. In comparison to state-of-the-art performance on the binary classification test, we achieved (97% accuracy, 97% sensitivity, 98.3% specificity, and 0.9455, Quadratic Weighted Kappa score (QWK)). Additionally, for severity grading, Our network demonstrated high proficiency (82% accuracy-0.888 (QWK)). The suggested approach makes a substantial contribution by accurately classifying the degree of diabetic retinopathy severity while requiring less time and spatial complexity, making it a promising contender for autonomous diagnosis.
International Journal of Advances in Soft Computing and its Applications, 2021
The number of people around the world who have diabetes is about 422 million. Diabetes seriously affects the blood vessels in the retina, a disease called diabetic retinopathy (DR). The ophthalmologist examines signs through fundus images, such microaneurysm, exudates and neovascularisation and determines the suitable treatment for patient based on the condition. Currently, doctors require a long time and professional skills to detect DR. This study aimed to implement artificial intelligence (AI) to resolve the lack of current methods. This study implemented AI for detecting and classifying DR. AI uses deep learning, such the attention mechanism algorithm and AlexNet architecture. The attention mechanism algorithm focuses on detecting the pathological area in the fundus images, and AlexNet is used to classify DR into five levels based on the pathological area. This study also compared AlexNet architecture with and without attention mechanism. We obtained 344 fundus images from the K...
Diabetic Retinopathy Detection and Grading using Deep learning
MJEER, 2022
One of the complications of diabetes disease is diabetic retinopathy (DR). Diabetic patients may suffer from total loss of sight. That's if it is not detected and medicated early enough. The early detection of DR is very important during funds screening on a regular basis. Detection and grading of DR are difficult because most fundus images suffer from undersaturation and noise. This paper proposes a new enhancement process as a solution to the poor quality of fundus images. It also proposes two architectures for convolutional neural network (CNN) models. The first one is the binary classifier of DR images into normal and abnormal. The second CNN architecture to classify the severity grades of DR. In this study, we also utilized different pre-trained convolutional neural network models to show the impact on the performance of the use of transfer learning from pre-trained CNN models vs newly defined architectures. The pre-trained CNN models and the two new proposed CNN models are tested using Messidor1, Messidor2, and Kaggle EyePACS datasets. The proposed binary classifier model results in F1-scores of 0.9387, 0.9629, and 0.9430 on the Messidor-1, Messidor-2, and EyePACS datasets, respectively. The proposed second model classifies the five grades with an F1-score of 0.9133, 0.9226, and 0.9393 on the Messidor1, Messidor2, and Kaggle EyePACS datasets, respectively. The new proposed CNN model proved its reliability and efficiency in detecting DR and classifying severity grades of DR in fundus images. Preprocessing techniques enhanced the performance by 10.83% of accuracy and 0.13037 in AUC using the binary model.
AUTOMATIC DETECTION OF SEVERITY GRADING IN DIABETIC RETINOPATHY USING CONVOLUTIONAL NEURAL NETWORK
IRJET, 2023
The primary reason for middle-aged people's eyesight is age is diabetic retinopathy (DR). Early identification of the development of diabetic retinopathy can be very beneficial for clinical treatment. Although several different feature extraction various strategies have been put forth, and the classification job for retinal images is still tedious and time-consuming even for those trained clinicians. Hence, primary screening of DR is to avoid vision loss, it is advised that diabetic patients have this procedure performed at least once a year. Recently, deep convolutional neural networks have manifested superior performance in image classification compared to previous handcrafted feature-based image classification methods. As a result, a Random forest classifier has been developed that can distinguish the intricate elements required for classification, such as micro-aneurysms, exudate, and hemorrhages on the retina, and then automatically deliver a diagnosis without human input. Last but not least, a CNN-based automated DR screening approach for retinal pictures is suggested. This method displays the different phases of DR (Mild, Moderate, and Severe) as well as its attention map for the region that is most affected. It also reduces the workload of ophthalmologists. Thus the proposed system of CNN classifier gives a significant improvement in terms of speed and accuracy when compared to previous methods.
Predicting Severity of Diabetic Retinopathy using Deep Learning Models
International Research Journal on Advanced Science Hub, 2021
This paper presents deep learning models for the classification of Diabetic Retinopathy (DR) grades. The goal of this research is to find and create a deep learning model that will help us identify the image with high accuracy into one of the five phases of the DR as no DR, mild, moderate, severe, and proliferative DR.The whole work is developed using four steps. The first, using Ben Graham's pre-possessing form, the fundus images were pre-processed. Secondly, in order to train the models, the preprocessed images are contributed to the deep learning algorithm. The third,deep learning models such as Deep CNN, Dense Net, and Group 19 Visual Geometry (VGG19) are developed to predict the severity of the DR. The APTOS Blindness Detection dataset is used to train the proposed deep learning models. Since the data set is imbalanced in nature, the issue of training bias contributes to it. Therefore, at the time of training the models, class weight technique is used to eliminate the training bias problem. In the case of DR grading structures, the proposed deep learning models work well. The Dense Net has been found to work better than the other two models.
Design and development of a deep learning based application for detecting diabetic retinopathy
2019
Diabetic retinopathy (DR), a complication of diabetes, is one of the leading causes of blindness globally. Since early detection of DR can reduce the chance of vision loss significantly, regular retinal screening of diabetic patients is an essential prerequisite. However, due to inefficient manual detection as well as lack of resources and ophthalmologists, early detection of DR is severely hindered. Moreover, subtle differences among different severity levels and the presence of small anatomical components make the task of identification very challenging. The objective of this study is to develop a robust diagnostic system through integration of state-of-theart deep learning techniques for automated DR severity detection. We used the concept of deep Convolutional Neural Networks (CNNs), which have revolutionized different branches of computer vision including medical imaging. Our deep network is trained on the largest publicly available Kaggle data set using our very own novel loss...
Diabetic retinopathy classification using deep convolutional neural network
Indonesian Journal of Electrical Engineering and Computer Science, 2021
Diabetic retinopathy (DR) is a diabetic impairment that affects the eyes and if not treated could lead to permanent vision impairment. Traditionally, Ophthalmologists perform diagnosis of DR by checking for existence and any seriousness of some subtle features in the fundus images. This process is not very efficient as it takes a lot of time and resources. DR testing of all the patients, a lot of which are undiagnosed or untreated, is a big task due to the inefficiency of the traditional method. This paper was written with the aim to propose a classification system based on an efficient deep convolution neural network (DCNN) model which is computationally efficient. Amongst other supervised algorithms involved, proposed solution is to find a way to efficiently classify the fundus images into 5 different levels of severity. Application of segmentation after the pre-processing and then use of deep convolutional neural networks on the dataset results in a high accuracy of 91.52%. The r...
Deep Learning-based Diabetic Retinopathy Detection for automated health assessment
Diabetic Retinopathy(DR) is a type of chronic microvascular diabetes complication that deteriorates the human vision system and leads to a patient towards complete blindness. DR is a common condition among people with diabetes, affecting an estimated 93 million people. Early detection of DR is a vital strategy to alleviate massive vision impairment. In this research, we have experienced DR patients and tried to determine whether they are DR affected or not. We present a lightweight and quick detection system for classifying DR over fundus images of two sets: healthy retina or DR-affected retina. After pre-processing, we produced the dataset annotations needed for model training. Next, we add the MobilenetV2 to Depthwise Separable Convolution (DSC) level to compute the representative collection of data. The proposed system is trained in a standard approach for further classification using a cross-entropy loss function. We have performed comprehensive testing on a dataset of 3662 fundus images from the Kaggle competition. The proposed model performs rapid detection with a high-level accuracy of 97.82%. It also obtained 95% sensitivity and 96% specificity on the test dataset. We have evaluated our method with the existing contemporary approaches to highlight its robustness based on DR classification. We have achieved incredible results over other traditional methods based on computational complexity and training time. The proposed early DR detection system will be a promising tool for improving the management and treatment of diabetic patients.
DIABETIC RETINOPATHY DETECTION USING DEEP LEARNING
IAEME PUBLICATION, 2020
Diabetic Retinopathy (DR) is an ailment that arises in patients suffering from Diabetes for more than 2 decades. This causes blindness due to elevated blood glucose levels. The condition can be cured when it is diagnosed at an early stage. The currently available methods to diagnose Diabetic Retinopathy is Fluorescein Angiography, which is slow and is not available to masses. In this paper we propose a novel solution by incorporating Convolutional Neural Networks (CNN) to detect the presence of Diabetic Retinopathy using the color fundus image of the patient. We have built a Neural Network architecture consisting of two stages of CNN where the first stage detects the presence of DR and the second stage classifies it into four stages. With this architecture, we have achieved an accuracy of 90.32%. The dataset we used to train the Neural Network was taken from a Kaggle competition that consisted of 3,662 color fundus images.